Abstract
We use unsupervised machine learning techniques of natural language processing to measure the semantic relatedness of firms’ narrative risk disclosures (NRD) in the “risk factor” section of their annual reports. We find strong return predictability among firms with overlapping NRD. This momentum effect is distinct from return predictability due to existing factors such as industry momentum, complexity of firm operations, and size-based information diffusion. A long-short strategy based on the NRD relatedness yields a monthly alpha of 70 basis points. Further analyses show that this lead-lag return effect is stronger among firms with lower investor attention and higher illiquidity. Our findings highlight important asset pricing implications of textual relatedness in NRD that investors seem to be inattentive towards